MediVoice

MediVoice is a voice-based mobile app that helps visually impaired people manage their medications on their own. With this app, users can take a picture of a drug label and ask questions by speaking. The app then reads the label, understands the question, and gives a clear voice response. It tells users things like the name of the medicine

---

🚀 Project Overview

1. Image Capture & OCR

   • Automatically find and process the latest image files in the labels directory.
   • Uses EasyOCR to extract raw text from medication labels.
   • Cleans and normalizes OCR output (lowercase, remove non-alphanumerics, drop stopwords).

2. Speech-to-Text

   • Scans your Recordings/ directory for new audio files (.wav, .mp3, .m4a, etc.).
   • Converts to WAV (if needed) and transcribes using OpenAI Whisper.

3. LLM Inference

   • Combines clean OCR text and user question into a natural prompt.
   • Sends the prompt to a hosted Hugging Face Inference endpoint (no local model download).
   • Returns a clear, conversational answer or a polite fallback if data is insufficient.

4. Workflow Automation

   • main.py ties together all pipelines, making it easy to run everything in one command.

---

📂 Repository Structure

MediVoice/
├── README.md           # This file
├── main.py             # Entry point: runs OCR, STT, and LLM inference
├── config.py           # Directory paths & environment overrides
├── requirements.txt    # Python dependencies
├── .env                # Store your API keys
└── scripts/
    ├── label_extraction.py  # OCR helpers: get_latest_image, extract_text, preprocess
    ├── speech_to_text.py    # Audio helpers: convert_to_wav, speech_to_text
    └── llm.py               # Inference helpers: build_prompt & generate_answer

---

🔧 Setup & Installation

1. Clone the repo

   git clone https://github.com/yourorg/MediVoice.git
   cd MediVoice

2. Install dependencies

   pip install --upgrade pip
   pip install -r requirements.txt

---

▶️ Usage

Run the full pipeline with:

python main.py

• OCR will process the newest label*.jpg in labels/.
• Speech-to-Text will transcribe the newest audio file in Recordings/.
• LLM Inference will generate a response using your Hugging Face hosted model.

---